Worksheet: Argumentation Mining

Identifying claims, premises, and argumentative relations
Course: Natural Language Annotation for Machine Learning Task Type: Structured annotation (components + relations)
Author: Jin Zhao

Background

Argumentation mining automatically identifies the structure of arguments in text: what claims are made, what evidence supports them, and how different arguments relate.

Claim: A statement that the author wants the reader to accept

Premise: A reason or justification offered in support of a claim

Evidence: Factual information (data, citations) supporting a premise

Argumentative Relations

RelationDescriptionExample
SupportPremise supports claimP: "Exercise reduces stress" → C: "People should exercise"
AttackPremise undermines claimP: "Exercise takes time" → C: "People should exercise"
RebuttalCounter-argument to attack"But the time investment pays health dividends"

Component Labels

CLAIM PREMISE EVIDENCE REBUTTAL

Part 1: Identifying Components

(1) Universities should require all students to take a public speaking course. (2) Communication skills are essential in virtually every career. (3) A 2023 LinkedIn survey found that 89% of employers rank communication as the most valued soft skill.

Question 1

Verify the component labels above and identify the relations:

Sentence (2) Sentence (1)

Sentence (3) Sentence (2)

Part 2: Implicit Claims

The city spent $50 million on the new stadium while three schools remain without air conditioning. Meanwhile, teacher salaries have been frozen for five years.

Question 2

What is the implicit claim being made?

Should implicit claims be annotated?

Part 3: Support vs. Attack

Claim: "Remote work should become the permanent standard for office jobs."

Classify each statement as supporting or attacking the claim:

Question 3
StatementRelation
"Studies show remote workers report higher job satisfaction."
"Remote work can lead to feelings of isolation and loneliness."
"Companies save on office space costs with remote workers."
"Collaboration and spontaneous creativity suffer without in-person interaction."
"The technology for effective remote work now exists."

Part 4: Complex Argument Structure

(A) We should ban plastic bags. (B) They harm marine life. (C) Over 100,000 marine animals die from plastic annually. (D) However, paper bags require more energy to produce. (E) But the long-term environmental cost of plastic is far greater. (F) Plastic takes 500 years to decompose.

Question 4

Label each component and map the relations:

SentenceComponent TypeRelates ToRelation Type
(A)
(B)
(C)
(D)
(E)
(F)

Part 5: Argument Quality

Argument 1: "We shouldn't trust scientists about climate change because they just want grant money."

Argument 2: "Climate change is real because 97% of climate scientists agree, based on peer-reviewed research analyzing temperature data from the past 150 years."

Question 5

Evaluate the quality of each argument:

DimensionArgument 1Argument 2
Evidence quality
Logical validity
Persuasiveness

Part 6: Annotate This Passage

Social media should be banned for children under 16. Young minds are particularly susceptible to the addictive design of these platforms. Research from Stanford shows that teen anxiety has increased 70% since smartphones became ubiquitous. Some argue that social media helps kids stay connected, but studies show that online interactions don't provide the same mental health benefits as face-to-face friendships. The potential for cyberbullying is another serious concern, with 37% of teens reporting online harassment.

Question 6

Identify all argumentative components and their relations:

Part 7: Group Discussion

Question 7

Compare your annotations with your group. Where did you disagree?

Part 8: Reflection

Question 8

Why is argumentation mining difficult?

Key Takeaway

Argumentation is structured reasoning, but structure is often implicit and subjective.

  • The same text can be interpreted as having different argument structures
  • Whether something supports or attacks depends on perspective
  • Implicit claims and premises are common in natural text
  • Quality evaluation requires both logical and domain expertise